SAFNWC/MSG Cloud type/height. Application for fog/low cloud situations 22 September 2011 Hervé LE GLEAU, Marcel DERRIEN Centre de météorologie Spatiale. Lannion Météo-France 1
Fog or low level clouds? It is irrealistic to claim that we can identify fog only using satellite data: we do not know if the cloud observed from satellite is reaching the ground. -> I will therefore not present fog mapping from satellite. -> Instead: I will first detail the cloud products extracted from MSG/SEVIRI satellite imagery using the NWCSAF software, concentrating on the fog or low level clouds category. I will then show an example of data fusion with NWCSAF cloud products to map fog risk. 2
Plan SAFNWC context Main features of SAFNWC/MSG cloud algorithms Cma cloud mask CT cloud type CTTH cloud top temperature and height Summary of validation results Illustration with fog/low cloud situations: (including example of automatic use for fog risk mapping) Outlook 3
SAFNWC context -SAFNWC delivers software to process data from MSG and polar platforms (METOP/NOAA). - 92 registered users, including 29 European NMS and 3 SAFs (OSISAF, CMSAF, LSASAF) -SAFNWC/MSG SW includes three cloud products (CMa, CT, CTTH) developed by Météo-France/Lannion -Detailed description of cloud algorithms and validation results available from www.nwcsaf.inm.org -SAFNC/MSG SW v2011 will be used during this presentation. 4
CMa algorithm: first step CMa First step: Clouds and snow are first detected in each pixel of the image using multispectral theshold techiques : Thresholds are computed using : o Atlas: height map land/sea mask o Climatological maps: SST continental visible reflectance o NWP short range forecast data (at MF, Arpege 1.5 deg used): surface temperature, integrated atmospheric precipitable water Thresholds tuned to radiometer s spectral characteristics with Radiative Transfer Models in cloud free conditions (6S,RTTOV). 5
Illustration of night-time low cloud identification T10.8μm T3.9μm T8.7μm T10.8μm 6
Illustration of night-time low cloud identification Low clouds T10.8μm T3.9μm T8.7μm T10.8μm 7
Illustration of daytime low cloud identification VIS 0.6μm T3.9μm-T10.8μm VIS 1.6μm 8
Illustration of daytime low cloud identification Low clouds VIS 0.6μm T3.9μm-T10.8μm VIS 1.6μm 9
Illustration of daytime low cloud identification Low clouds VIS 0.6μm T3.9μm-T10.8μm Snow VIS 1.6μm 10
CMa algorithm: second step CMa Second step: (only available since version v2009 (available to users in march)) Temporal analysis and region-growing technique are applied to detect low clouds at day-night transition and fast moving clouds: For fast moving clouds: detect T10.8μm changes within 15 minutes For low clouds in day-night transition: the areas, cloudy 1hour before, that have unchanged T10.8μm, T12.0μm and T8.7μm during last hour are said cloudy + spatial extension of these cloudy areas to adjacent areas having similar Vis06μm reflectance and T10.8μm 11
Illustration of improvement with temporal analysis 93 93 1h sooner 80 Cloud mask + temporal scheme superimposed on BRF 0.6 μm 12
CMa: decrease of false alarm over snow (night) The following problem has been reported by users: «At night during winter cold events, cloud-free snow-coved grounds may be wrongly classified as clouds». These wrong detection are due to any of three tests applied to T10.8μm, T3.9μm-T10.8μm or T8.7μm-T10.8μm An empirical approach has been applied to solve the problem (v2011): relax thresholds when cold snow-coved grounds are expected 13
CMa: decrease of false alarm over snow (night) Diagnose where strong nocturnal cooling may occur altitude < 1500m and (Ts nwp < 263K or Ts nwp < 268K and Snow occur. > 5) Relax three thresholds -T108thr -5.0K if T108thr 255K or T108thr -5.0K -.4(255.-T108thr) if T108thr < 255K -T87T108thr +0.4K if T108thr<250K -T39T18thr=MAX(T39T10.8thr, -0.5x T108thr+129.0) if 250K T108thr 255K or T39T18thr=-0.15x T108thr +41.5 if T108thr<250K Clear restoral when detected by T108thr test at any illumination, or Visible test or T39T108thr test at daytime or twilight: If t108thr < 250K and T7.3-T10.8> 0.5K 14
CMa: decrease of false alarm over snow (night) 15
CMa: decrease of false alarm over snow (night) 16
CT algorithm Cloudy pixels are classified according their radiative characteristics: Semi-transparent and fractional clouds are distinguished from low/medium/high clouds using spectral features: low T10.8μm-T12.0μm, low T8.7μm-T10.8μm high T10.8μm-T3.9μm (night), high R0.6μm (day) Low, mid-level and high clouds are then separated by comparing their T10.8μm to combination of NWP forecast temperature at various pressure levels [850, 700, 500 hpa and at tropopause levels]. 17
CT: decrease low/mid-level cloud confusion Low clouds may be wrongly classified as mid-level clouds in the presence of a thermal inversion. Two approaches are used to minimise the confusion: mid-level clouds are reclassified as low clouds if T10.8μm- WV73μm is «large» mid-level clouds are reclassified as low clouds if a low level thermal inversion is detected in the NWP fields input by the user and if T8.7μm-T10.8μm is lower than a threshold (decreasing with viewing angles) The improvement is illustrated over central Europe on 21 st December 2007 18
CT: decrease low/mid-level cloud confusion V2010 V2009 19
CTTH algorithm Vertical temperature & humidity profile forecast by NWP needed TOA radiances from the top of overcast opaque clouds put at various pressure levels are simulated with RTTOV (NWP vertical profiles are temporally interpolated to each slot) Cloud top pressure is first extracted using RTTOV simulated radiances; Method depending on cloud type. Cloud top temperature & height are derived from their pressure (using vertical temperature & humidity profile forecast by NWP). 20
CTTH algorithm For opaque clouds (known from CT) The cloud top pressure corresponds to the best fit between the simulated and measured 10.8μm radiances For semi-transparent clouds : Derived from a window channel 10.8μm and a sounding channel (13.4μm, 7.3μm or 6.2μm) For broken low clouds No technique has yet been implemented. 21
Illustration of opaque clouds cloud top pressure retrieval 22
Measured brightness temperature Illustration of opaque clouds cloud top pressure retrieval 23
Measured brightness temperature Retrieved cloud top pressure Illustration of opaque clouds cloud top pressure retrieval 24
Illustration of opaque clouds cloud top pressure retrieval in case thermal inversion 25
Illustration of opaque clouds cloud top pressure retrieval in case thermal inversion Measured brightness temperature 26
Retrieved cloud top pressure Illustration of opaque clouds cloud top pressure retrieval in case thermal inversion Measured brightness temperature 27
Illustration of opaque clouds cloud top pressure retrieval in case thermal inversion 28
Illustration of opaque clouds cloud top pressure retrieval in case thermal inversion Measured brightness temperature 29
(in case of dry air between 850 and 600) (ie, relative humidity lower than 30%) Illustration of opaque clouds cloud top pressure retrieval in case thermal inversion Measured brightness temperature 30
(in case of dry air between 850 and 600) (ie, relative humidity lower than 30%) Retrieved cloud top pressure Illustration of opaque clouds cloud top pressure retrieval in case thermal inversion Measured brightness temperature 31
CTTH pressure example 32
Summary of CMa validation with SYNOP 500 manned continental station over Europe from 10th Decembre 2010 to 21st March 2011 Following cloudiness are compared: SEVIRI: average cloudiness in a 5x5 target SYNOP: total observed cloudiness POD (%) FAR (%) Daytime 98.1 1.7 Night-time 95.7 6.2 Twilight 95.5 1.9 High FAR partly due to error in night-time human cloud observation. Lower POD mainly due to low cloud underdetection 33
Summary of CT visual inspection (related to low cloud) Stability of CT classifier to illumination Low clouds may be occasionaly undetected at night-time (especially oceanic rather warm Sc advected above not too cold ground) Low cloud identication at day-night transition: mainly solved in v2009. Over land, tendency to classify low clouds as mid-level (in case strong thermal inversion): mainly solved in v2010 Night-time confusion of snow as clouds: mainly solved in v2011. 34
Validation of low cloud CTTH with ground-based radar September 2003-October 2004 Following cloud top height are compared: derived from cloud radar (95Ghz) from SIRTA (LMD, near Paris) computed from SEVIRI (CTH_SEVIRI - CTH_radar > 0) = SEVIRI CTH overestimation Cloud type Mean (km) STD (km) Low opaque 0.28 0.96 Low opaque if thermal inversion observed in NWP 0.17 0.62 35
14/02/08: documented by Maria Putsay (Hungary) on Eumetsat web Image gallery 14/02/2008 01h25 36
14/02/08: documented by Maria Putsay (Hungary) on Eumetsat web Image gallery 14/02/2008 01h00 37
14/02/08: documented by Maria Putsay (Hungary) on Eumetsat web Image gallery 14/02/2008 10h40 38
14/02/08: documented by Maria Putsay (Hungary) on Eumetsat web Image gallery 14/02/2008 12h00 39
Exemple of automated use for fog risk mapping A combined use of: SAFNWC/MSG CT, rain accumulation and NWP analysis (air humidity (2m), wind (10m)) 40
Outlook Future upgrade of NWCSAF SW: -inclusion of microphysical product: --cloud phase, --effective radius size, --optical thickness, --water/ice water path -ready for MTG: --more channels and better spatial resolution -long-term development: separation between stratiform and cumuliform clouds: --for low clouds : the separation of small cumulus and stratiform clouds will be useful for fog risk estimation. 41
For further information For more information on NWCSAF: www.nwcsaf.org For further information of NWCSAF software (freely available): mafernandeza@aemet.es Near-real time NWCSAF products can be visualized on: www.nwcsaf.org 42